Rank

Model

average

V1 1 benchmark

V2 1 benchmark

V4 1 benchmark

IT 2 benchmarks

behavior 1 benchmark

engineering 1 benchmark

Deng2009-top1 v1

1 CORnet-S
Kubilius et al., 2018
.417 .294 .294 .242 .242 .581 .581 .423 .541 .305 .545 .545 .747 .747
2 vgg-19
Simonyan et al., 2014
.408 .347 .347 .341 .341 .610 .610 .248 .496 X .494 .494 .711 .711
3 resnet-50-robust
Santurkar et al., 2019
.408 .378 .378 .365 .365 .537 .537 .243 .486 X .515 .515
4 resnet-101_v1
He et al., 2015
.407 .266 .266 .341 .341 .590 .590 .274 .549 X .561 .561 .764 .764
5 vgg-16
Simonyan et al., 2014
.406 .355 .355 .336 .336 .620 .620 .259 .518 X .461 .461 .715 .715
6 resnet-152_v1
He et al., 2015
.405 .282 .282 .338 .338 .598 .598 .277 .553 X .533 .533 .768 .768
7 resnet-101_v2
He et al., 2015
.404 .274 .274 .332 .332 .599 .599 .263 .527 X .555 .555 .774 .774
8 resnet50-SIN_IN
Geirhos et al., 2019
.404 .282 .282 .324 .324 .599 .599 .276 .552 X .541 .541 .746 .746
9 densenet-169
Huang et al., 2016
.404 .281 .281 .322 .322 .601 .601 .274 .548 X .543 .543 .759 .759
10 densenet-201
Huang et al., 2016
.402 .277 .277 .325 .325 .599 .599 .273 .545 X .537 .537 .772 .772
11 resnet-50-pytorch
He et al., 2015
.399 .289 .289 .317 .317 .600 .600 .259 .518 X .528 .528 .752 .752
12 resnet-50_v1
He et al., 2015
.398 .274 .274 .317 .317 .594 .594 .278 .555 X .526 .526 .752 .752
13 resnet50-SIN_IN_IN
Geirhos et al., 2019
.397 .275 .275 .321 .321 .596 .596 .273 .545 X .523 .523 .767 .767
14 resnet-152_v2
He et al., 2015
.397 .274 .274 .326 .326 .591 .591 .266 .532 X .528 .528 .778 .778
15 resnet-50_v2
He et al., 2015
.396 .270 .270 .323 .323 .596 .596 .260 .520 X .531 .531 .756 .756
16 densenet-121
Huang et al., 2016
.396 .277 .277 .306 .306 .595 .595 .267 .533 X .535 .535 .745 .745
17 resnext101_32x32d_wsl
Mahajan et al., 2018
.396 .267 .267 .289 .289 .574 .574 .254 .507 X .594 .594 .854 .854
18 mobilenet_v1_1.0_160
Howard et al., 2017
.393 .290 .290 .332 .332 .588 .588 .275 .549 X .480 .480 .680 .680
19 resnext101_32x8d_wsl
Mahajan et al., 2018
.392 .271 .271 .312 .312 .586 .586 .241 .481 X .551 .551 .842 .842
20 inception_v2
Szegedy et al., 2015
.392 .284 .284 .313 .313 .587 .587 .270 .539 X .505 .505 .739 .739
21 resnet-18
He et al., 2015
.390 .274 .274 .302 .302 .583 .583 .266 .531 X .524 .524 .698 .698
22 mobilenet_v2_1.0_224
Howard et al., 2017
.389 .245 .245 .331 .331 .573 .573 .273 .546 X .521 .521 .718 .718
23 mobilenet_v2_0.75_224
Howard et al., 2017
.388 .236 .236 .316 .316 .586 .586 .268 .535 X .533 .533 .698 .698
24 efficientnet-b0
Tan et al., 2019
.387 .215 .215 .317 .317 .556 .556 .274 .547 X .573 .573
25 fixres_resnext101_32x48d_wsl
Touvron et al., 2019
.387 .246 .246 .288 .288 .582 .582 .257 .513 X .561 .561 .863 .863
25 resnext101_32x48d_wsl
Mahajan et al., 2018
.387 .246 .246 .288 .288 .582 .582 .257 .513 X .561 .561 .822 .822
27 mobilenet_v2_1.3_224
Howard et al., 2017
.386 .253 .253 .332 .332 .575 .575 .271 .543 X .500 .500 .744 .744
28 resnet50-SIN
Geirhos et al., 2019
.386 .300 .300 .333 .333 .580 .580 .267 .534 X .448 .448 .602 .602
29 pnasnet_large
Liu et al., 2017
.385 .264 .264 .305 .305 .578 .578 .263 .526 X .515 .515 .829 .829
30 AT_efficientnet-b7
Xie et al., 2020
.385 .276 .276 .308 .308 .583 .583 .281 .562 X .475 .475
31 mobilenet_v2_0.75_192
Howard et al., 2017
.384 .245 .245 .306 .306 .573 .573 .275 .550 X .524 .524 .687 .687
32 mobilenet_v2_1.4_224
Howard et al., 2017
.384 .257 .257 .321 .321 .566 .566 .277 .554 X .500 .500 .750 .750
33 inception_v1
Szegedy et al., 2014
.384 .259 .259 .311 .311 .589 .589 .244 .488 X .518 .518 .698 .698
34 xception
Chollet et al., 2016
.384 .245 .245 .306 .306 .610 .610 .249 .498 X .508 .508 .790 .790
35 AT_efficientnet-b4
Xie et al., 2020
.383 .246 .246 .339 .339 .549 .549 .279 .559 X .503 .503
36 mobilenet_v2_0.75_160
Howard et al., 2017
.383 .278 .278 .316 .316 .573 .573 .273 .547 X .473 .473 .664 .664
37 inception_v4
Szegedy et al., 2016
.382 .238 .238 .299 .299 .574 .574 .263 .526 X .537 .537 .802 .802
38 resnext101_32x16d_wsl
Mahajan et al., 2018
.382 .263 .263 .302 .302 .587 .587 .250 .499 X .509 .509 .851 .851
39 inception_resnet_v2
Szegedy et al., 2016
.381 .233 .233 .319 .319 .583 .583 .272 .543 X .499 .499 .804 .804
40 efficientnet-b6
Tan et al., 2019
.381 .263 .263 .295 .295 .563 .563 .271 .541 X .513 .513
41 efficientnet-b2
Tan et al., 2019
.380 .213 .213 .317 .317 .569 .569 .273 .547 X .526 .526
42 nasnet_large
Zoph et al., 2017
.380 .282 .282 .291 .291 .585 .585 .270 .541 X .470 .470 .827 .827
43 mobilenet_v1_1.0_224
Howard et al., 2017
.380 .223 .223 .341 .341 .560 .560 .273 .546 X .502 .502 .709 .709
44 efficientnet-b4
Tan et al., 2019
.379 .228 .228 .286 .286 .575 .575 .272 .543 X .535 .535
45 inception_v3
Szegedy et al., 2015
.379 .241 .241 .307 .307 .596 .596 .273 .545 X .477 .477 .780 .780
46 mobilenet_v2_1.0_192
Howard et al., 2017
.377 .216 .216 .322 .322 .572 .572 .273 .547 X .503 .503 .707 .707
47 mobilenet_v2_1.0_160
Howard et al., 2017
.376 .239 .239 .322 .322 .570 .570 .275 .550 X .472 .472 .688 .688
48 mobilenet_v2_0.5_192
Howard et al., 2017
.375 .263 .263 .329 .329 .566 .566 .264 .529 X .454 .454 .639 .639
49 mobilenet_v2_0.5_224
Howard et al., 2017
.372 .229 .229 .308 .308 .569 .569 .266 .533 X .488 .488 .654 .654
50 mobilenet_v1_0.75_224
Howard et al., 2017
.372 .223 .223 .336 .336 .558 .558 .267 .535 X .477 .477 .684 .684
51 AT_efficientnet-b2
Xie et al., 2020
.372 .248 .248 .295 .295 .563 .563 .275 .550 X .480 .480
52 resnet-34
He et al., 2015
.372 .230 .230 .286 .286 .560 .560 .237 .474 X .546 .546 .733 .733
53 AT_efficientnet-b0
Xie et al., 2020
.371 .238 .238 .334 .334 .570 .570 .267 .534 X .447 .447
54 mobilenet_v1_0.5_224
Howard et al., 2017
.370 .221 .221 .340 .340 .555 .555 .260 .521 X .474 .474 .633 .633
55 mobilenet_v1_1.0_192
Howard et al., 2017
.370 .235 .235 .329 .329 .548 .548 .271 .543 X .466 .466 .700 .700
56 mobilenet_v2_0.75_128
Howard et al., 2017
.369 .237 .237 .320 .320 .553 .553 .271 .541 X .464 .464 .632 .632
57 alexnet
None
.368 .316 .316 .353 .353 .550 .550 .254 .508 X .370 .370 .577 .577
58 mobilenet_v1_1.0_128
Howard et al., 2017
.368 .254 .254 .325 .325 .557 .557 .267 .535 X .437 .437 .652 .652
59 mobilenet_v1_0.75_128
Howard et al., 2017
.368 .267 .267 .330 .330 .564 .564 .252 .505 X .425 .425 .621 .621
60 mobilenet_v2_0.5_160
Howard et al., 2017
.368 .258 .258 .305 .305 .562 .562 .264 .528 X .448 .448 .610 .610
61 mobilenet_v2_1.0_128
Howard et al., 2017
.368 .252 .252 .303 .303 .569 .569 .267 .534 X .447 .447 .653 .653
62 mobilenet_v1_0.5_192
Howard et al., 2017
.367 .220 .220 .337 .337 .566 .566 .260 .520 X .454 .454 .617 .617
63 mobilenet_v1_0.75_192
Howard et al., 2017
.367 .229 .229 .339 .339 .549 .549 .267 .535 X .449 .449 .672 .672
64 mobilenet_v2_0.35_192
Howard et al., 2017
.366 .264 .264 .301 .301 .568 .568 .259 .518 X .437 .437 .582 .582
65 mobilenet_v2_1.0_96
Howard et al., 2017
.363 .256 .256 .332 .332 .530 .530 .257 .514 X .443 .443 .603 .603
66 resnet18-supervised
He et al., 2015
.361 .276 .276 .281 .281 .539 .539 .263 .526 X .446 .446
67 mobilenet_v1_0.5_160
Howard et al., 2017
.361 .265 .265 .320 .320 .557 .557 .252 .503 X .410 .410 .591 .591
68 mobilenet_v2_0.35_160
Howard et al., 2017
.359 .269 .269 .292 .292 .554 .554 .259 .517 X .424 .424 .557 .557
69 mobilenet_v1_0.75_160
Howard et al., 2017
.359 .213 .213 .346 .346 .558 .558 .264 .529 X .413 .413 .653 .653
70 mobilenet_v2_0.35_224
Howard et al., 2017
.359 .215 .215 .296 .296 .554 .554 .253 .506 X .474 .474 .603 .603
71 mobilenet_v2_0.5_128
Howard et al., 2017
.358 .222 .222 .309 .309 .557 .557 .262 .525 X .440 .440 .577 .577
72 nasnet_mobile
Zoph et al., 2017
.357 .272 .272 .273 .273 .566 .566 .268 .536 X .406 .406 .740 .740
73 mobilenet_v2_0.75_96
Howard et al., 2017
.350 .208 .208 .305 .305 .527 .527 .258 .516 X .451 .451 .588 .588
74 squeezenet1_0
Iandola et al., 2016
.341 .304 .304 .320 .320 .591 .591 .229 .459 X .263 .263 .575 .575
75 mobilenet_v1_0.5_128
Howard et al., 2017
.341 .245 .245 .304 .304 .550 .550 .234 .467 X .373 .373 .563 .563
76 squeezenet1_1
Iandola et al., 2016
.336 .265 .265 .311 .311 .582 .582 .229 .457 X .291 .291 .575 .575
77 mobilenet_v2_0.35_128
Howard et al., 2017
.333 .245 .245 .289 .289 .530 .530 .235 .470 X .367 .367 .508 .508
78 mobilenet_v2_0.5_96
Howard et al., 2017
.331 .266 .266 .278 .278 .501 .501 .239 .479 X .370 .370 .512 .512
79 mobilenet_v1_0.25_224
Howard et al., 2017
.327 .231 .231 .296 .296 .538 .538 .240 .480 X .333 .333 .498 .498
80 mobilenet_v1_0.25_192
Howard et al., 2017
.323 .208 .208 .318 .318 .517 .517 .226 .451 X .344 .344 .477 .477
81 CORnet-Z
Kubilius et al., 2018
.322 .298 .298 .182 .182 .553 .553 .223 .447 X .356 .356 .470 .470
82 resnet18-local_aggregation
Zhuang et al., 2019
.314 .253 .253 .308 .308 .563 .563 .268 .536 X .177 .177
83 mobilenet_v1_0.25_160
Howard et al., 2017
.312 .198 .198 .293 .293 .509 .509 .229 .457 X .330 .330 .455 .455
84 bagnet9
Brendel et al., 2019
.307 .215 .215 .260 .260 .550 .550 .200 .401 X .307 .307 .260 .260
85 mobilenet_v2_0.35_96
Howard et al., 2017
.303 .183 .183 .249 .249 .501 .501 .230 .460 X .351 .351 .455 .455
86 mobilenet_v1_0.25_128
Howard et al., 2017
.302 .262 .262 .238 .238 .513 .513 .213 .425 X .286 .286 .415 .415
87 vggface
Parkhi et al., 2015
.301 .358 .358 .339 .339 .555 .555 .176 .351 X .078 .078
88 resnet18-contrastive_multiview
Zhuang et al., 2020
.293 .258 .258 .265 .265 .551 .551 .231 .461 X .161 .161
89 resnet18-instance_recognition
Wu et al., 2018
.292 .267 .267 .294 .294 .548 .548 .261 .522 X .090 .090
90 resnet18-colorization
Zhuang et al., 2020
.273 .269 .269 .265 .265 .568 .568 .205 .410 X .060 .060
91 resnet18-deepcluster
Zhuang et al., 2020
.272 .258 .258 .306 .306 .545 .545 .253 .506 X .000 X
92 resnet18-relative_position
Zhuang et al., 2020
.262 .278 .278 .302 .302 .544 .544 .194 .388 X -0.006 -0.006
93 resnet18-depth_prediction
Zhuang et al., 2020
.260 .285 .285 .246 .246 .509 .509 .158 .315 X .102 .102
94 dcgan
None
.242 .316 .316 .226 .226 .432 .432 .214 .214 .023 .023
95 resnet18-contrastive_predictive
Zhuang et al., 2020
.236 .247 .247 .263 .263 .497 .497 .163 .325 X .010 .010
96 prednet
Zhuang et al., 2020
.222 .224 .224 .234 .234 .503 .503 .138 .275 X .014 .014
97 resnet18-autoencoder
Zhuang et al., 2020
.218 .298 .298 .165 .165 .438 .438 .103 .207 X .083 .083
98 pixels
None
.030 .053 .053 .003 .003 .068 .068 .008 .015 X .020 .020
Model scores on brain benchmarks. Hover over model name to see layer commitments. The more green and bright a cell, the better the model's score. Scores are ceiled, hover the benchmark to see ceilings.

About

The Brain-Score platform aims to yield strong computational models of the ventral stream. We enable researchers to quickly get a sense of how their model scores against standardized brain benchmarks on multiple dimensions and facilitate comparisons to other state-of-the-art models. At the same time, new brain data can quickly be tested against a wide range of models to determine how well existing models explain the data.

Brain-Score is organized by the Brain-Score team in collaboration with researchers and labs worldwide. We are working towards an easy-to-use platform where a model can easily be submitted to yield its scores on a range of brain benchmarks and new benchmarks can be incorporated to challenge the models.

This quantified approach lets us keep track of how close our models are to the brain on a range of experiments (data) using different evaluation techniques (metrics). For more details, please refer to the technical paper and the perspective paper.

Compare

x vs y.
Hover over dots to reveal model details. Scrolling zooms in and out. Click the dropdowns to change x-/y-axis data.

Participate

Challenge the data: Submit a model

If you would like to score a model, please log in here.

Challenge the models: Submit data

If you have neural or behavioral recordings that you would like models to compete on, please get in touch with us to submit data.

Change the evaluation: Submit a metric

If you have an idea for a different way of comparing brain and machine, please send in a pull request.

Citation

If you use Brain-Score in your work, please cite Brain-Score: Which Artificial Neural Network for Object Recognition is most Brain-Like? (technical) and Integrative Benchmarking to Advance Neurally Mechanistic Models of Human Intelligence (perspective) as well as the respective benchmark sources.
@article{SchrimpfKubilius2018BrainScore,
  title={Brain-Score: Which Artificial Neural Network for Object Recognition is most Brain-Like?},
  author={Martin Schrimpf and Jonas Kubilius and Ha Hong and Najib J. Majaj and Rishi Rajalingham and Elias B. Issa and Kohitij Kar and Pouya Bashivan and Jonathan Prescott-Roy and Franziska Geiger and Kailyn Schmidt and Daniel L. K. Yamins and James J. DiCarlo},
  journal={bioRxiv preprint},
  year={2018},
  url={https://www.biorxiv.org/content/10.1101/407007v2}
}

@article{Schrimpf2020integrative,
  title={Integrative Benchmarking to Advance Neurally Mechanistic Models of Human Intelligence},
  author={Schrimpf, Martin and Kubilius, Jonas and Lee, Michael J and Murty, N Apurva Ratan and Ajemian, Robert and DiCarlo, James J},
  journal={Neuron},
  year={2020},
  url={https://www.cell.com/neuron/fulltext/S0896-6273(20)30605-X}
}